The Neural Basis of Birdsong Syntax

Many complex actions such as speaking and playing music consist of sequences of action units strung together according to syntactic rules. How such rules are implemented in the brain is a critical problem for understanding the neural basis of complex behaviors. Songbirds are ideal model systems for tackling this problem. Birdsong consists of sequences of stereotypical syllables. The syntax of the syllable sequences ranges from simple to complex in different species. Many species such as Bengalese finch sing songs with variable sequences with probabilistic transitions between the syllables. Neural circuits underlying the singing behavior are well characterized, and are accessible to experimental manipulations and detailed computational modeling. Experimental and modeling works have established that syllables are encoded in unidirectional chain networks of projection neurons in the songbird premotor nucleus HVC (used as a proper name). Spike propagation along a chain network drives downstream neurons and produces a specific syllable. Computational models suggest that the song syntax can be embedded in the connection patterns of the syllable encoding chains. Allowed transitions from a syllable to other syllables are encoded by connecting the chain networks associated with the syllables into branched patterns. Through a winner-take-all mechanism enforced through the local inhibitory circuit and noise, a single chain is selected to propagate the spikes at a branching point, producing a probabilistic syllable transition. Probabilistic state transition models inspired by the network models can accurately describe the statistical properties of observed Bengalese finch song sequences, supporting the network mechanism of birdsong syntax.